-
-
Save eli-oat/5f8b94e08cd8817c580e590f8bf70c82 to your computer and use it in GitHub Desktop.
A minimal Forth compiler in ANSI C
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
/******************************************************************************* | |
* | |
* A minimal Forth compiler in C | |
* By Leif Bruder <[email protected]> http://defineanswer42.wordpress.com | |
* Release 2014-04-04 | |
* | |
* Based on Richard W.M. Jones' excellent Jonesforth sources/tutorial | |
* | |
* PUBLIC DOMAIN | |
* | |
* I, the copyright holder of this work, hereby release it into the public | |
* domain. This applies worldwide. In case this is not legally possible, I grant | |
* any entity the right to use this work for any purpose, without any conditions, | |
* unless such conditions are required by law. | |
* | |
*******************************************************************************/ | |
/* Only a single include here; I'll define everything on the fly to keep | |
* dependencies as low as possible. In this file, the only C standard functions | |
* used are getchar, putchar and the EOF value. */ | |
#include <stdio.h> | |
/* Base cell data types. Use short/long on most systems for 16 bit cells. */ | |
/* Experiment here if necessary. */ | |
#define CELL_BASE_TYPE int | |
#define DOUBLE_CELL_BASE_TYPE long | |
/* Basic memory configuration */ | |
#define MEM_SIZE 65536 /* main memory size in bytes */ | |
#define STACK_SIZE 192 /* cells reserved for the stack */ | |
#define RSTACK_SIZE 64 /* cells reserved for the return stack */ | |
#define INPUT_LINE_SIZE 32 /* bytes reserved for the WORD buffer */ | |
/******************************************************************************/ | |
/* Our basic data types */ | |
typedef CELL_BASE_TYPE scell; | |
typedef DOUBLE_CELL_BASE_TYPE dscell; | |
typedef unsigned CELL_BASE_TYPE cell; | |
typedef unsigned DOUBLE_CELL_BASE_TYPE dcell; | |
typedef unsigned char byte; | |
#define CELL_SIZE sizeof(cell) | |
#define DCELL_SIZE sizeof(dcell) | |
/* A few constants that describe the memory layout of this implementation */ | |
#define LATEST_POSITION INPUT_LINE_SIZE | |
#define HERE_POSITION (LATEST_POSITION + CELL_SIZE) | |
#define BASE_POSITION (HERE_POSITION + CELL_SIZE) | |
#define STATE_POSITION (BASE_POSITION + CELL_SIZE) | |
#define STACK_POSITION (STATE_POSITION + CELL_SIZE) | |
#define RSTACK_POSITION (STACK_POSITION + STACK_SIZE * CELL_SIZE) | |
#define HERE_START (RSTACK_POSITION + RSTACK_SIZE * CELL_SIZE) | |
#define MAX_BUILTIN_ID 71 | |
/* Flags and masks for the dictionary */ | |
#define FLAG_IMMEDIATE 0x80 | |
#define FLAG_HIDDEN 0x40 | |
#define MASK_NAMELENGTH 0x1F | |
/* This is the main memory to be used by this Forth. There will be no malloc | |
* in this file. */ | |
byte memory[MEM_SIZE]; | |
/* Pointers to Forth variables stored inside the main memory array */ | |
cell *latest; | |
cell *here; | |
cell *base; | |
cell *state; | |
cell *sp; | |
cell *stack; | |
cell *rsp; | |
cell *rstack; | |
/* A few helper variables for the compiler */ | |
int exitReq; | |
int errorFlag; | |
cell next; | |
cell lastIp; | |
cell quit_address; | |
cell commandAddress; | |
cell maxBuiltinAddress; | |
/* The TIB, stored outside the main memory array for now */ | |
char lineBuffer[128]; | |
int charsInLineBuffer = 0; | |
int positionInLineBuffer = 0; | |
/* A basic setup for defining builtins. This Forth uses impossibly low | |
* adresses as IDs for the builtins so we can define builtins as | |
* standard C functions. Slower but easier to port. */ | |
#define BUILTIN(id, name, c_name, flags) const int c_name##_id=id; const char* c_name##_name=name; const byte c_name##_flags=flags; void c_name() | |
#define ADD_BUILTIN(c_name) addBuiltin(c_name##_id, c_name##_name, c_name##_flags, c_name) | |
typedef void(*builtin)(); | |
builtin builtins[MAX_BUILTIN_ID] = { 0 }; | |
/* This is our initialization script containing all the words we define in | |
* Forth for convenience. Focus is on simplicity, not speed. Partly copied from | |
* Jonesforth (see top of file). */ | |
char *initscript_pos; | |
const char *initScript = | |
": DECIMAL 10 BASE ! ;\n" | |
": HEX 16 BASE ! ;\n" | |
": OCTAL 8 BASE ! ;\n" | |
": 2DUP OVER OVER ;\n" | |
": 2DROP DROP DROP ;\n" | |
": NIP SWAP DROP ;\n" | |
": 2NIP 2SWAP 2DROP ;\n" | |
": TUCK SWAP OVER ;\n" | |
": / /MOD NIP ;\n" | |
": MOD /MOD DROP ;\n" | |
": BL 32 ;\n" | |
": CR 10 EMIT ;\n" | |
": SPACE BL EMIT ;\n" | |
": NEGATE 0 SWAP - ;\n" | |
": DNEGATE 0. 2SWAP D- ;\n" | |
": CELLS CELL * ;\n" | |
": ALLOT HERE @ + HERE ! ;\n" | |
": TRUE -1 ;\n" | |
": FALSE 0 ;\n" | |
": 0= 0 = ;\n" | |
": 0< 0 < ;\n" | |
": 0> 0 > ;\n" | |
": <> = 0= ;\n" | |
": <= > 0= ;\n" | |
": >= < 0= ;\n" | |
": 0<= 0 <= ;\n" | |
": 0>= 0 >= ;\n" | |
": 1+ 1 + ;\n" | |
": 1- 1 - ;\n" | |
": 2+ 2 + ;\n" | |
": 2- 2 - ;\n" | |
": 2/ 2 / ;\n" | |
": 2* 2 * ;\n" | |
": D2/ 2. D/ ;\n" | |
": +! DUP @ ROT + SWAP ! ;\n" | |
": [COMPILE] WORD FIND >CFA , ; IMMEDIATE\n" | |
": [CHAR] key ' LIT , , ; IMMEDIATE\n" | |
": RECURSE LATEST @ >CFA , ; IMMEDIATE\n" | |
": DOCOL 0 ;\n" | |
": CONSTANT CREATE DOCOL , ' LIT , , ' EXIT , ;\n" | |
": 2CONSTANT SWAP CREATE DOCOL , ' LIT , , ' LIT , , ' EXIT , ;\n" | |
": VARIABLE HERE @ CELL ALLOT CREATE DOCOL , ' LIT , , ' EXIT , ;\n" /* TODO: Allot AFTER the code, not before */ | |
": 2VARIABLE HERE @ 2 CELLS ALLOT CREATE DOCOL , ' LIT , , ' EXIT , ;\n" /* TODO: Allot AFTER the code, not before */ | |
": IF ' 0BRANCH , HERE @ 0 , ; IMMEDIATE\n" | |
": THEN DUP HERE @ SWAP - SWAP ! ; IMMEDIATE\n" | |
": ELSE ' BRANCH , HERE @ 0 , SWAP DUP HERE @ SWAP - SWAP ! ; IMMEDIATE\n" | |
": BEGIN HERE @ ; IMMEDIATE\n" | |
": UNTIL ' 0BRANCH , HERE @ - , ; IMMEDIATE\n" | |
": AGAIN ' BRANCH , HERE @ - , ; IMMEDIATE\n" | |
": WHILE ' 0BRANCH , HERE @ 0 , ; IMMEDIATE\n" | |
": REPEAT ' BRANCH , SWAP HERE @ - , DUP HERE @ SWAP - SWAP ! ; IMMEDIATE\n" | |
": UNLESS ' 0= , [COMPILE] IF ; IMMEDIATE\n" | |
": DO HERE @ ' SWAP , ' >R , ' >R , ; IMMEDIATE\n" | |
": LOOP ' R> , ' R> , ' SWAP , ' 1+ , ' 2DUP , ' = , ' 0BRANCH , HERE @ - , ' 2DROP , ; IMMEDIATE\n" | |
": +LOOP ' R> , ' R> , ' SWAP , ' ROT , ' + , ' 2DUP , ' <= , ' 0BRANCH , HERE @ - , ' 2DROP , ; IMMEDIATE\n" | |
": I ' R@ , ; IMMEDIATE\n" | |
": SPACES DUP 0> IF 0 DO SPACE LOOP ELSE DROP THEN ;\n" | |
": ABS DUP 0< IF NEGATE THEN ;\n" | |
": DABS 2DUP 0. D< IF DNEGATE THEN ;\n" | |
": .DIGIT DUP 9 > IF 55 ELSE 48 THEN + EMIT ;\n" | |
": .SIGN DUP 0< IF 45 EMIT NEGATE THEN ;\n" /* BUG: 10000000000... will be shown wrong */ | |
": .POS BASE @ /MOD ?DUP IF RECURSE THEN .DIGIT ;\n" | |
": . .SIGN DUP IF .POS ELSE .DIGIT THEN ;\n" | |
": COUNTPOS SWAP 1 + SWAP BASE @ / ?DUP IF RECURSE THEN ;\n" | |
": DIGITS DUP 0< IF 1 ELSE 0 THEN SWAP COUNTPOS ;\n" | |
": .R OVER DIGITS - SPACES . ;\n" | |
": . . SPACE ;\n" | |
": ? @ . ;\n" | |
": .S DSP@ BEGIN DUP S0@ > WHILE DUP ? CELL - REPEAT DROP ;\n" | |
": TYPE 0 DO DUP C@ EMIT 1 + LOOP DROP ;\n" | |
": ALIGN BEGIN HERE @ CELL MOD WHILE 0 C, REPEAT ;\n" | |
": s\" ' LITSTRING , HERE @ 0 , BEGIN KEY DUP 34 <> WHILE C, REPEAT DROP DUP HERE @ SWAP - CELL - SWAP ! ALIGN ; IMMEDIATE\n" | |
": .\" [COMPILE] s\" ' TYPE , ; IMMEDIATE\n" | |
": ( BEGIN KEY [CHAR] ) = UNTIL ; IMMEDIATE\n" | |
": COUNT DUP 1+ SWAP C@ ;\n" | |
": MIN 2DUP < IF DROP ELSE NIP THEN ;\n" | |
": MAX 2DUP > IF DROP ELSE NIP THEN ;\n" | |
": D0= OR 0= ;\n" | |
": DMIN 2OVER 2OVER D< IF 2DROP ELSE 2NIP THEN ;\n" | |
": DMAX 2OVER 2OVER D> IF 2DROP ELSE 2NIP THEN ;\n" | |
; | |
/******************************************************************************/ | |
/* The primary data output function. This is the place to change if you want | |
* to e.g. output data on a microcontroller via a serial interface. */ | |
void putkey(char c) | |
{ | |
putchar(c); | |
} | |
/* The primary data input function. This is where you place the code to e.g. | |
* read from a serial line. */ | |
int llkey() | |
{ | |
if (*initscript_pos) return *(initscript_pos++); | |
return getchar(); | |
} | |
/* Anything waiting in the keyboard buffer? */ | |
int keyWaiting() | |
{ | |
return positionInLineBuffer < charsInLineBuffer ? -1 : 0; | |
} | |
/* Line buffered character input. We're duplicating the functionality of the | |
* stdio library here to make the code easier to port to other input sources */ | |
int getkey() | |
{ | |
int c; | |
if (keyWaiting()) | |
return lineBuffer[positionInLineBuffer++]; | |
charsInLineBuffer = 0; | |
while ((c = llkey()) != EOF) | |
{ | |
if (charsInLineBuffer == sizeof(lineBuffer)) break; | |
lineBuffer[charsInLineBuffer++] = c; | |
if (c == '\n') break; | |
} | |
positionInLineBuffer = 1; | |
return lineBuffer[0]; | |
} | |
/* C string output */ | |
void tell(const char *str) | |
{ | |
while (*str) | |
putkey(*str++); | |
} | |
/* The basic (data) stack operations */ | |
cell pop() | |
{ | |
if (*sp == 1) | |
{ | |
tell("? Stack underflow\n"); | |
errorFlag = 1; | |
return 0; | |
} | |
return stack[--(*sp)]; | |
} | |
cell tos() | |
{ | |
if (*sp == 1) | |
{ | |
tell("? Stack underflow\n"); | |
errorFlag = 1; | |
return 0; | |
} | |
return stack[(*sp)-1]; | |
} | |
void push(cell data) | |
{ | |
if (*sp >= STACK_SIZE) | |
{ | |
tell("? Stack overflow\n"); | |
errorFlag = 1; | |
return; | |
} | |
stack[(*sp)++] = data; | |
} | |
dcell dpop() | |
{ | |
cell tmp[2]; | |
tmp[1] = pop(); | |
tmp[0] = pop(); | |
return *((dcell*)tmp); | |
} | |
void dpush(dcell data) | |
{ | |
cell tmp[2]; | |
*((dcell*)tmp) = data; | |
push(tmp[0]); | |
push(tmp[1]); | |
} | |
/* The basic return stack operations */ | |
cell rpop() | |
{ | |
if (*rsp == 1) | |
{ | |
tell("? RStack underflow\n"); | |
errorFlag = 1; | |
return 0; | |
} | |
return rstack[--(*rsp)]; | |
} | |
void rpush(cell data) | |
{ | |
if (*rsp >= RSTACK_SIZE) | |
{ | |
tell("? RStack overflow\n"); | |
errorFlag = 1; | |
return; | |
} | |
rstack[(*rsp)++] = data; | |
} | |
/* Secure memory access */ | |
cell readMem(cell address) | |
{ | |
if (address > MEM_SIZE) | |
{ | |
tell("Internal error in readMem: Invalid addres\n"); | |
errorFlag = 1; | |
return 0; | |
} | |
return *((cell*)(memory + address)); | |
} | |
void writeMem(cell address, cell value) | |
{ | |
if (address > MEM_SIZE) | |
{ | |
tell("Internal error in writeMem: Invalid address\n"); | |
errorFlag = 1; | |
return; | |
} | |
*((cell*)(memory + address)) = value; | |
} | |
/* Reading a word into the input line buffer */ | |
byte readWord() | |
{ | |
char *line = (char*)memory; | |
byte len = 0; | |
int c; | |
while ((c = getkey()) != EOF) | |
{ | |
if (c == ' ') continue; | |
if (c == '\n') continue; | |
if (c != '\\') break; | |
while ((c = getkey()) != EOF) | |
if (c == '\n') | |
break; | |
} | |
while (c != ' ' && c != '\n' && c != EOF) | |
{ | |
if (len >= (INPUT_LINE_SIZE - 1)) | |
break; | |
line[++len] = c; | |
c = getkey(); | |
} | |
line[0] = len; | |
return len; | |
} | |
/* toupper() clone so we don't have to pull in ctype.h */ | |
char up(char c) | |
{ | |
return (c >= 'a' && c <= 'z') ? c - 'a' + 'A' : c; | |
} | |
/* Dictionary lookup */ | |
cell findWord(cell address, cell len) | |
{ | |
cell ret = *latest; | |
char *name = (char*)&memory[address]; | |
cell i; | |
int found; | |
for (ret = *latest; ret; ret = readMem(ret)) | |
{ | |
if ((memory[ret + CELL_SIZE] & MASK_NAMELENGTH) != len) continue; | |
if (memory[ret + CELL_SIZE] & FLAG_HIDDEN) continue; | |
found = 1; | |
for (i = 0; i < len; i++) | |
{ | |
if (up(memory[ret + i + 1 + CELL_SIZE]) != up(name[i])) | |
{ | |
found = 0; | |
break; | |
} | |
} | |
if (found) break; | |
} | |
return ret; | |
} | |
/* Basic number parsing, base <= 36 only atm */ | |
void parseNumber(byte *word, cell len, dcell *number, cell *notRead, byte *isDouble) | |
{ | |
int negative = 0; | |
cell i; | |
char c; | |
cell current; | |
*number = 0; | |
*isDouble = 0; | |
if (len == 0) | |
{ | |
*notRead = 0; | |
return; | |
} | |
if (word[0] == '-') | |
{ | |
negative = 1; | |
len--; | |
word++; | |
} | |
else if (word[0] == '+') | |
{ | |
len--; | |
word++; | |
} | |
for (i = 0; i < len; i++) | |
{ | |
c = *word; | |
word++; | |
if (c == '.') { *isDouble = 1; continue; } | |
else if (c >= '0' && c <= '9') current = c - '0'; | |
else if (c >= 'A' && c <= 'Z') current = 10 + c - 'A'; | |
else if (c >= 'a' && c <= 'z') current = 10 + c - 'a'; | |
else break; | |
if (current >= *base) break; | |
*number = *number * *base + current; | |
} | |
*notRead = len - i; | |
if (negative) *number = (-((scell)*number)); | |
} | |
/******************************************************************************* | |
* | |
* Builtin definitions | |
* | |
*******************************************************************************/ | |
BUILTIN(0, "RUNDOCOL", docol, 0) | |
{ | |
rpush(lastIp); | |
next = commandAddress + CELL_SIZE; | |
} | |
/* The first few builtins are very simple, not need to waste vertical space here */ | |
BUILTIN( 1, "CELL", doCellSize, 0) { push(CELL_SIZE); } | |
BUILTIN( 2, "@", memRead, 0) { push(readMem(pop())); } | |
BUILTIN( 3, "C@", memReadByte, 0) { push(memory[pop()]); } | |
BUILTIN( 4, "KEY", key, 0) { push(getkey()); } | |
BUILTIN( 5, "EMIT", emit, 0) { putkey(pop() & 255); } | |
BUILTIN( 6, "DROP", drop, 0) { pop(); } | |
BUILTIN( 7, "EXIT", doExit, 0) { next = rpop(); } | |
BUILTIN( 8, "BYE", bye, 0) { exitReq = 1; } | |
BUILTIN( 9, "LATEST", doLatest, 0) { push(LATEST_POSITION); } | |
BUILTIN(10, "HERE", doHere, 0) { push(HERE_POSITION); } | |
BUILTIN(11, "BASE", doBase, 0) { push(BASE_POSITION); } | |
BUILTIN(12, "STATE", doState, 0) { push(STATE_POSITION); } | |
BUILTIN(13, "[", gotoInterpreter, FLAG_IMMEDIATE) { *state = 0; } | |
BUILTIN(14, "]", gotoCompiler, 0) { *state = 1; } | |
BUILTIN(15, "HIDE", hide, 0) { memory[*latest + CELL_SIZE] ^= FLAG_HIDDEN; } | |
BUILTIN(16, "R>", rtos, 0) { push(rpop()); } | |
BUILTIN(17, ">R", stor, 0) { rpush(pop()); } | |
BUILTIN(18, "KEY?", key_p, 0) { push(keyWaiting()); } | |
BUILTIN(19, "BRANCH", branch, 0) { next += readMem(next); } | |
BUILTIN(20, "0BRANCH", zbranch, 0) { next += pop() ? CELL_SIZE : readMem(next); } | |
BUILTIN(21, "IMMEDIATE", toggleImmediate, FLAG_IMMEDIATE) { memory[*latest + CELL_SIZE] ^= FLAG_IMMEDIATE; } | |
BUILTIN(22, "FREE", doFree, 0) { push(MEM_SIZE - *here); } | |
BUILTIN(23, "S0@", s0_r, 0) { push(STACK_POSITION + CELL_SIZE); } | |
BUILTIN(24, "DSP@", dsp_r, 0) { push(STACK_POSITION + *sp * CELL_SIZE); } | |
BUILTIN(25, "NOT", not, 0) { push(~pop()); } | |
BUILTIN(26, "DUP", dup, 0) { push(tos()); } | |
BUILTIN(27, "!", memWrite, 0) | |
{ | |
cell address = pop(); | |
cell value = pop(); | |
writeMem(address, value); | |
} | |
BUILTIN(28, "C!", memWriteByte, 0) | |
{ | |
cell address = pop(); | |
cell value = pop(); | |
memory[address] = value & 255; | |
} | |
BUILTIN(29, "SWAP", swap, 0) | |
{ | |
cell a = pop(); | |
cell b = pop(); | |
push(a); | |
push(b); | |
} | |
BUILTIN(30, "OVER", over, 0) | |
{ | |
cell a = pop(); | |
cell b = tos(); | |
push(a); | |
push(b); | |
} | |
BUILTIN(31, ",", comma, 0) | |
{ | |
push(*here); | |
memWrite(); | |
*here += CELL_SIZE; | |
} | |
BUILTIN(32, "C,", commaByte, 0) | |
{ | |
push(*here); | |
memWriteByte(); | |
*here += sizeof(byte); | |
} | |
BUILTIN(33, "WORD", word, 0) | |
{ | |
byte len = readWord(); | |
push(1); | |
push(len); | |
} | |
BUILTIN(34, "FIND", find, 0) | |
{ | |
cell len = pop(); | |
cell address = pop(); | |
cell ret = findWord(address, len); | |
push(ret); | |
} | |
cell getCfa(cell address) | |
{ | |
byte len = (memory[address + CELL_SIZE] & MASK_NAMELENGTH) + 1; | |
while ((len & (CELL_SIZE-1)) != 0) len++; | |
return address + CELL_SIZE + len; | |
} | |
BUILTIN(35, ">CFA", cfa, 0) | |
{ | |
cell address = pop(); | |
cell ret = getCfa(address); | |
if (ret < maxBuiltinAddress) | |
push(readMem(ret)); | |
else | |
push(ret); | |
} | |
BUILTIN(36, "NUMBER", number, 0) | |
{ | |
dcell num; | |
cell notRead; | |
byte isDouble; | |
cell len = pop(); | |
byte* address = &memory[pop()]; | |
parseNumber(address, len, &num, ¬Read, &isDouble); | |
if (isDouble) dpush(num); else push((cell)num); | |
push(notRead); | |
} | |
BUILTIN(37, "LIT", lit, 0) | |
{ | |
push(readMem(next)); | |
next += CELL_SIZE; | |
} | |
/* Outer and inner interpreter, TODO split up */ | |
BUILTIN(38, "QUIT", quit, 0) | |
{ | |
cell address; | |
dcell number; | |
cell notRead; | |
cell command; | |
int i; | |
byte isDouble; | |
cell tmp[2]; | |
int immediate; | |
for (exitReq = 0; exitReq == 0;) | |
{ | |
lastIp = next = quit_address; | |
errorFlag = 0; | |
word(); | |
find(); | |
address = pop(); | |
if (address) | |
{ | |
immediate = (memory[address + CELL_SIZE] & FLAG_IMMEDIATE); | |
commandAddress = getCfa(address); | |
command = readMem(commandAddress); | |
if (*state && !immediate) | |
{ | |
if (command < MAX_BUILTIN_ID && command != docol_id) | |
push(command); | |
else | |
push(commandAddress); | |
comma(); | |
} | |
else | |
{ | |
while (!errorFlag && !exitReq) | |
{ | |
if (command == quit_id) break; | |
else if (command < MAX_BUILTIN_ID) builtins[command](); | |
else | |
{ | |
lastIp = next; | |
next = command; | |
} | |
commandAddress = next; | |
command = readMem(commandAddress); | |
next += CELL_SIZE; | |
} | |
} | |
} | |
else | |
{ | |
parseNumber(&memory[1], memory[0], &number, ¬Read, &isDouble); | |
if (notRead) | |
{ | |
tell("Unknown word: "); | |
for (i=0; i<memory[0]; i++) | |
putkey(memory[i+1]); | |
putkey('\n'); | |
*sp = *rsp = 1; | |
continue; | |
} | |
else | |
{ | |
if (*state) | |
{ | |
*((dcell*)tmp) = number; | |
push(lit_id); | |
comma(); | |
if (isDouble) | |
{ | |
push(tmp[0]); | |
comma(); | |
push(lit_id); | |
comma(); | |
push(tmp[1]); | |
comma(); | |
} | |
else | |
{ | |
push((cell)number); | |
comma(); | |
} | |
} | |
else | |
{ | |
if (isDouble) dpush(number); else push((cell)number); | |
} | |
} | |
} | |
if (errorFlag) | |
*sp = *rsp = 1; | |
else if (!keyWaiting() && !(*initscript_pos)) | |
tell(" OK\n"); | |
} | |
} | |
BUILTIN(39, "+", plus, 0) | |
{ | |
scell n1 = pop(); | |
scell n2 = pop(); | |
push(n1 + n2); | |
} | |
BUILTIN(40, "-", minus, 0) | |
{ | |
scell n1 = pop(); | |
scell n2 = pop(); | |
push(n2 - n1); | |
} | |
BUILTIN(41, "*", mul, 0) | |
{ | |
scell n1 = pop(); | |
scell n2 = pop(); | |
push(n1 * n2); | |
} | |
BUILTIN(42, "/MOD", divmod, 0) | |
{ | |
scell n1 = pop(); | |
scell n2 = pop(); | |
push(n2 % n1); | |
push(n2 / n1); | |
} | |
BUILTIN(43, "ROT", rot, 0) | |
{ | |
cell a = pop(); | |
cell b = pop(); | |
cell c = pop(); | |
push(b); | |
push(a); | |
push(c); | |
} | |
void createWord(const char* name, byte len, byte flags); | |
BUILTIN(44, "CREATE", doCreate, 0) | |
{ | |
byte len; | |
cell address; | |
word(); | |
len = pop() & 255; | |
address = pop(); | |
createWord((char*)&memory[address], len, 0); | |
} | |
BUILTIN(45, ":", colon, 0) | |
{ | |
doCreate(); | |
push(docol_id); | |
comma(); | |
hide(); | |
*state = 1; | |
} | |
BUILTIN(46, ";", semicolon, FLAG_IMMEDIATE) | |
{ | |
push(doExit_id); | |
comma(); | |
hide(); | |
*state = 0; | |
} | |
BUILTIN(47, "R@", rget, 0) | |
{ | |
cell tmp = rpop(); | |
rpush(tmp); | |
push(tmp); | |
} | |
BUILTIN(48, "J", doJ, 0) | |
{ | |
cell tmp1 = rpop(); | |
cell tmp2 = rpop(); | |
cell tmp3 = rpop(); | |
rpush(tmp3); | |
rpush(tmp2); | |
rpush(tmp1); | |
push(tmp3); | |
} | |
BUILTIN(49, "'", tick, FLAG_IMMEDIATE) | |
{ | |
word(); | |
find(); | |
cfa(); | |
if (*state) | |
{ | |
push(lit_id); | |
comma(); | |
comma(); | |
} | |
} | |
BUILTIN(50, "=", equals, 0) | |
{ | |
cell a1 = pop(); | |
cell a2 = pop(); | |
push(a2 == a1 ? -1 : 0); | |
} | |
BUILTIN(51, "<", smaller, 0) | |
{ | |
scell a1 = pop(); | |
scell a2 = pop(); | |
push(a2 < a1 ? -1 : 0); | |
} | |
BUILTIN(52, ">", larger, 0) | |
{ | |
scell a1 = pop(); | |
scell a2 = pop(); | |
push(a2 > a1 ? -1 : 0); | |
} | |
BUILTIN(53, "AND", doAnd, 0) | |
{ | |
cell a1 = pop(); | |
cell a2 = pop(); | |
push(a2 & a1); | |
} | |
BUILTIN(54, "OR", doOr, 0) | |
{ | |
cell a1 = pop(); | |
cell a2 = pop(); | |
push(a2 | a1); | |
} | |
BUILTIN(55, "?DUP", p_dup, 0) | |
{ | |
cell a = tos(); | |
if (a) push(a); | |
} | |
BUILTIN(56, "LITSTRING", litstring, 0) | |
{ | |
cell length = readMem(next); | |
next += CELL_SIZE; | |
push(next); | |
push(length); | |
next += length; | |
while (next & (CELL_SIZE-1)) | |
next++; | |
} | |
BUILTIN(57, "XOR", xor, 0) | |
{ | |
cell a = pop(); | |
cell b = pop(); | |
push(a ^ b); | |
} | |
BUILTIN(58, "*/", timesDivide, 0) | |
{ | |
cell n3 = pop(); | |
dcell n2 = pop(); | |
dcell n1 = pop(); | |
dcell r = (n1 * n2) / n3; | |
push((cell)r); | |
if ((cell)r != r) | |
{ | |
tell("Arithmetic overflow\n"); | |
errorFlag = 1; | |
} | |
} | |
BUILTIN(59, "*/MOD", timesDivideMod, 0) | |
{ | |
cell n3 = pop(); | |
dcell n2 = pop(); | |
dcell n1 = pop(); | |
dcell r = (n1 * n2) / n3; | |
dcell m = (n1 * n2) % n3; | |
push((cell)m); | |
push((cell)r); | |
if ((cell)r != r) | |
{ | |
tell("Arithmetic overflow\n"); | |
errorFlag = 1; | |
} | |
} | |
BUILTIN(60, "D=", dequals, 0) | |
{ | |
dcell a1 = dpop(); | |
dcell a2 = dpop(); | |
push(a2 == a1 ? -1 : 0); | |
} | |
BUILTIN(61, "D<", dsmaller, 0) | |
{ | |
dscell a1 = dpop(); | |
dscell a2 = dpop(); | |
push(a2 < a1 ? -1 : 0); | |
} | |
BUILTIN(62, "D>", dlarger, 0) | |
{ | |
dscell a1 = dpop(); | |
dscell a2 = dpop(); | |
push(a2 > a1 ? -1 : 0); | |
} | |
BUILTIN(63, "DU<", dusmaller, 0) | |
{ | |
dcell a1 = dpop(); | |
dcell a2 = dpop(); | |
push(a2 < a1 ? -1 : 0); | |
} | |
BUILTIN(64, "D+", dplus, 0) | |
{ | |
dscell n1 = dpop(); | |
dscell n2 = dpop(); | |
dpush(n1 + n2); | |
} | |
BUILTIN(65, "D-", dminus, 0) | |
{ | |
dscell n1 = dpop(); | |
dscell n2 = dpop(); | |
dpush(n2 - n1); | |
} | |
BUILTIN(66, "D*", dmul, 0) | |
{ | |
dscell n1 = dpop(); | |
dscell n2 = dpop(); | |
dpush(n1 * n2); | |
} | |
BUILTIN(67, "D/", ddiv, 0) | |
{ | |
dscell n1 = dpop(); | |
dscell n2 = dpop(); | |
dpush(n2 / n1); | |
} | |
BUILTIN(68, "2SWAP", dswap, 0) | |
{ | |
dcell a = dpop(); | |
dcell b = dpop(); | |
dpush(a); | |
dpush(b); | |
} | |
BUILTIN(69, "2OVER", dover, 0) | |
{ | |
dcell a = dpop(); | |
dcell b = dpop(); | |
dpush(b); | |
dpush(a); | |
dpush(b); | |
} | |
BUILTIN(70, "2ROT", drot, 0) | |
{ | |
dcell a = dpop(); | |
dcell b = dpop(); | |
dcell c = dpop(); | |
dpush(b); | |
dpush(a); | |
dpush(c); | |
} | |
/******************************************************************************* | |
* | |
* Loose ends | |
* | |
*******************************************************************************/ | |
/* Create a word in the dictionary */ | |
void createWord(const char* name, byte len, byte flags) | |
{ | |
cell newLatest = *here; | |
push(*latest); | |
comma(); | |
push(len | flags); | |
commaByte(); | |
while (len--) | |
{ | |
push(*name); | |
commaByte(); | |
name++; | |
} | |
while (*here & (CELL_SIZE-1)) | |
{ | |
push(0); | |
commaByte(); | |
} | |
*latest = newLatest; | |
} | |
/* A simple strlen clone so we don't have to pull in string.h */ | |
byte slen(const char *str) | |
{ | |
byte ret = 0; | |
while (*str++) ret++; | |
return ret; | |
} | |
/* Add a builtin to the dictionary */ | |
void addBuiltin(cell code, const char* name, const byte flags, builtin f) | |
{ | |
if (errorFlag) return; | |
if (code >= MAX_BUILTIN_ID) | |
{ | |
tell("Error adding builtin "); | |
tell(name); | |
tell(": Out of builtin IDs\n"); | |
errorFlag = 1; | |
return; | |
} | |
if (builtins[code] != 0) | |
{ | |
tell("Error adding builtin "); | |
tell(name); | |
tell(": ID given twice\n"); | |
errorFlag = 1; | |
return; | |
} | |
builtins[code] = f; | |
createWord(name, slen(name), flags); | |
push(code); | |
comma(); | |
push(doExit_id); | |
comma(); | |
} | |
/* Program setup and jump to outer interpreter */ | |
int main() | |
{ | |
errorFlag = 0; | |
if (DCELL_SIZE != 2*CELL_SIZE) | |
{ | |
tell("Configuration error: DCELL_SIZE != 2*CELL_SIZE\n"); | |
return 1; | |
} | |
state = (cell*)&memory[STATE_POSITION]; | |
base = (cell*)&memory[BASE_POSITION]; | |
latest = (cell*)&memory[LATEST_POSITION]; | |
here = (cell*)&memory[HERE_POSITION]; | |
sp = (cell*)&memory[STACK_POSITION]; | |
stack = (cell*)&memory[STACK_POSITION + CELL_SIZE]; | |
rsp = (cell*)&memory[RSTACK_POSITION]; | |
rstack = (cell*)&memory[RSTACK_POSITION + CELL_SIZE]; | |
*sp = *rsp = 1; | |
*state = 0; | |
*base = 10; | |
*latest = 0; | |
*here = HERE_START; | |
ADD_BUILTIN(docol); | |
ADD_BUILTIN(doCellSize); | |
ADD_BUILTIN(memRead); | |
ADD_BUILTIN(memWrite); | |
ADD_BUILTIN(memReadByte); | |
ADD_BUILTIN(memWriteByte); | |
ADD_BUILTIN(key); | |
ADD_BUILTIN(emit); | |
ADD_BUILTIN(swap); | |
ADD_BUILTIN(dup); | |
ADD_BUILTIN(drop); | |
ADD_BUILTIN(over); | |
ADD_BUILTIN(comma); | |
ADD_BUILTIN(commaByte); | |
ADD_BUILTIN(word); | |
ADD_BUILTIN(find); | |
ADD_BUILTIN(cfa); | |
ADD_BUILTIN(doExit); | |
ADD_BUILTIN(quit); | |
quit_address = getCfa(*latest); | |
ADD_BUILTIN(number); | |
ADD_BUILTIN(bye); | |
ADD_BUILTIN(doLatest); | |
ADD_BUILTIN(doHere); | |
ADD_BUILTIN(doBase); | |
ADD_BUILTIN(doState); | |
ADD_BUILTIN(plus); | |
ADD_BUILTIN(minus); | |
ADD_BUILTIN(mul); | |
ADD_BUILTIN(divmod); | |
ADD_BUILTIN(rot); | |
ADD_BUILTIN(gotoInterpreter); | |
ADD_BUILTIN(gotoCompiler); | |
ADD_BUILTIN(doCreate); | |
ADD_BUILTIN(hide); | |
ADD_BUILTIN(lit); | |
ADD_BUILTIN(colon); | |
ADD_BUILTIN(semicolon); | |
ADD_BUILTIN(rtos); | |
ADD_BUILTIN(stor); | |
ADD_BUILTIN(rget); | |
ADD_BUILTIN(doJ); | |
ADD_BUILTIN(tick); | |
ADD_BUILTIN(key_p); | |
ADD_BUILTIN(equals); | |
ADD_BUILTIN(smaller); | |
ADD_BUILTIN(larger); | |
ADD_BUILTIN(doAnd); | |
ADD_BUILTIN(doOr); | |
ADD_BUILTIN(branch); | |
ADD_BUILTIN(zbranch); | |
ADD_BUILTIN(toggleImmediate); | |
ADD_BUILTIN(doFree); | |
ADD_BUILTIN(p_dup); | |
ADD_BUILTIN(s0_r); | |
ADD_BUILTIN(dsp_r); | |
ADD_BUILTIN(litstring); | |
ADD_BUILTIN(not); | |
ADD_BUILTIN(xor); | |
ADD_BUILTIN(timesDivide); | |
ADD_BUILTIN(timesDivideMod); | |
ADD_BUILTIN(dequals); | |
ADD_BUILTIN(dsmaller); | |
ADD_BUILTIN(dlarger); | |
ADD_BUILTIN(dusmaller); | |
ADD_BUILTIN(dplus); | |
ADD_BUILTIN(dminus); | |
ADD_BUILTIN(dmul); | |
ADD_BUILTIN(ddiv); | |
ADD_BUILTIN(dswap); | |
ADD_BUILTIN(dover); | |
ADD_BUILTIN(drot); | |
maxBuiltinAddress = (*here) - 1; | |
if (errorFlag) return 1; | |
initscript_pos = (char*)initScript; | |
quit(); | |
return 0; | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment